Design And Implementation Of Wearable Lower Limb Motion Intelligent Detection System Based On Machine Learning

Lower limb dyskinesia refers to the phenomenon that voluntary movement is not controlled by the will,mainly manifested as tremor,difficulty in starting,lower limb weakness,etc.,mostly due to diseases,mental disorders or trauma and other reasons,these symptoms can seriously affect the daily life of patients.The body tends to deteriorate with age,making movement disorders more difficult to detect and preventing patients from receiving effective treatment early in their symptoms.Therefore,early detection of sports symptoms in order to timely medical treatment has become the first problem to be solved.The main assessment method for dyskinesia is the motor scale,but this method is highly subjective and more dependent on the experience of clinicians,and is not suitable for the autonomous real-time detection of users in daily life,so it is in urgent need of an accurate and objective detection method for lower limb dyskinesia.While the inertial sensor can accurately and objectively collect human movement data,STM32 SCM has low power consumption and good real-time performance,which is mostly used for embedded system implementation of wearable devices.Therefore,this paper mainly studies the wearable intelligent detection system for lower limb movement disorders,so as to remind people to receive treatment as early as possible in daily life and assist clinicians to make diagnosis.The main research content of this paper:(1)Data collection: In this paper,inertial sensors were used to collect the movement data of the subjects,and three experimental movements were independently designed for the detection of lower limb movement disorders by referring to the scale used for the clinical evaluation of movement disorders of Parkinson’s disease.The three experimental movements were horizontal stepping on the ground,heel point and high leg lift respectively.The inertial sensor was worn on the outside of the subject’s ankle,and the triaxial acceleration and triaxial angular velocity were collected for the three experimental movements.(2)Data processing and feature extraction: The pre-processing of the collected motion data is mainly divided into two parts: manual screening and low-pass filtering.According to the experimental scheme,the data set was divided into eight groups,six groups were divided according to movements and limbs,and the left lower limb feature set and the right lower limb feature set were added.The time domain features and frequency domain features were extracted,and the energy features and logarithmic energy features were extracted for the triaxial acceleration.A total of 134 features were extracted for each single body movement.The feature set realized the preliminary feature selection through the feature significance analysis of the experimental group and the control group,and then carried out different feature selection for a variety of machine learning classifiers.Finally,each data set left 30 important features for subsequent experiments.The training set,verification set and test set are divided according to3:1:1 ratio.(3)Model comparison experiment: Seven machine learning classification models mostly used in the field of motion evaluation were selected,model comparison experiment was designed,and the divided training set was used to train the models;The model parameters were optimized by traversal method.The average area under the curve was used to evaluate the model performance,and the single factor analysis of variance was carried out on the evaluation results of the model on the test set to determine that there were significant differences between the model results.By comparing the evaluation results of the model,a software model suitable for the implementation of the hardware system was obtained,namely the Gaussian kernel support vector machine model of the left lower limb heel movement,whose average area under the curve reached 96.5%.(4)Hardware implementation of evaluation model: overall design of hardware system;The main components are selected.STM32f103ZET6 is the microcontroller and JY901 chip is the sensor module.The hardware system simulation experiment was carried out by STM32 development board.The hardware realized the intelligent detection model for lower limb movement disorders,and optimized the process of motion solution,feature extraction and model recognition in the process of model realization.PCB plate making and hardware system of the equipment shell design;Finally,I completed the plate making and importing program,completed the embedded system for intelligent detection of lower limb movement disorders,and carried out hardware testing.